KDDE - Knowledge Discovery and Data Engineering

11/08/2026

📢 𝗡𝗲𝘄 𝗞𝗗𝗗𝗘 𝗽𝗮𝗽𝗲𝗿 𝗻𝗼𝘄 𝗼𝗻𝗹𝗶𝗻𝗲!

Can adversarial malware fool not only a detector, but also the explanation of its decision?

We are pleased to share our new paper:

𝗔𝗱𝘃𝗲𝗿𝘀𝗮𝗿𝗶𝗮𝗹 𝗠𝗮𝗹𝘄𝗮𝗿𝗲 𝗖𝗮𝗻 𝗕𝗲 𝗕𝗼𝘁𝗵 𝗘𝘃𝗮𝘀𝗶𝘃𝗲 𝗮𝗻𝗱 𝗗𝗲𝗰𝗲𝗶𝘃𝗶𝗻𝗴: 𝗮 𝗚𝗿𝗮𝗱𝗶𝗲𝗻𝘁-𝗯𝗮𝘀𝗲𝗱 𝗔𝘁𝘁𝗮𝗰𝗸 𝗔𝗴𝗮𝗶𝗻𝘀𝘁 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻 𝗮𝗻𝗱 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗶𝗻 𝗪𝗶𝗻𝗱𝗼𝘄𝘀 𝗣𝗘 𝗠𝗮𝗹𝘄𝗮𝗿𝗲 𝗗𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻

by Luca Lobascio, Giuseppina Andresini, Annalisa Appice, and Donato Malerba.

The paper introduces 𝗚𝗔𝗠𝗘𝟰𝗘𝗫𝗘, a gradient-based adversarial framework designed to pursue two goals at the same time: making Windows PE malware evade deep neural malware detectors and manipulating the corresponding explanations so that they resemble those of benign software.

This dual perspective — attacking both 𝗽𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻 and 𝗲𝘅𝗽𝗹𝗮𝗻𝗮𝘁𝗶𝗼𝗻 — raises an important challenge for trustworthy AI in cybersecurity: securing model decisions may not be enough if the explanations used to understand those decisions can also be deceived.

The preliminary evaluation investigates GAME4EXE against two deep learning malware detectors, MalConv and BBDNN, exploring classification evasion, transferability, and the ability to produce goodware-like explanations.

📍 2026 IEEE 11th European Symposium on Security and Privacy Workshops (EuroS&PW)
🔗 𝗥𝗲𝗮𝗱 𝘁𝗵𝗲 𝗽𝗮𝗽𝗲𝗿 𝗼𝗻 𝗜𝗘𝗘𝗘 𝗫𝗽𝗹𝗼𝗿𝗲
DOI: 10.1109/EuroSPW72509.2026.00038

Research supported by FAIR – Future AI Research, Spoke 6 – Symbiotic AI, and SERICS under the NRRP MUR programme funded by the European Union – NextGenerationEU.

03/08/2026

📢 𝐍𝐞𝐰 𝐎𝐩𝐞𝐧 𝐀𝐜𝐜𝐞𝐬𝐬 𝐩𝐮𝐛𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧 𝐟𝐫𝐨𝐦 𝐭𝐡𝐞 𝐊𝐃𝐃𝐄 𝐋𝐚𝐛!

We are pleased to announce the publication of:

𝐂𝐈𝐂𝐄𝐑𝐎𝐍𝐄: 𝐀 𝐧𝐚𝐭𝐮𝐫𝐚𝐥 𝐥𝐚𝐧𝐠𝐮𝐚𝐠𝐞-𝐛𝐚𝐬𝐞𝐝 𝐠𝐥𝐨𝐛𝐚𝐥 𝐚𝐩𝐩𝐫𝐨𝐚𝐜𝐡 𝐟𝐨𝐫 𝐨𝐛𝐣𝐞𝐜𝐭-𝐜𝐞𝐧𝐭𝐫𝐢𝐜 𝐏𝐫𝐞𝐝𝐢𝐜𝐭𝐢𝐯𝐞 𝐏𝐫𝐨𝐜𝐞𝐬𝐬 𝐌𝐨𝐧𝐢𝐭𝐨𝐫𝐢𝐧𝐠

by Vincenzo Pasquadibisceglie, 𝐀𝐧𝐧𝐚𝐥𝐢𝐬𝐚 𝐀𝐩𝐩𝐢𝐜𝐞, and Donato Malerba.

CICERONE introduces an original approach to 𝐨𝐛𝐣𝐞𝐜𝐭-𝐜𝐞𝐧𝐭𝐫𝐢𝐜 𝐏𝐫𝐞𝐝𝐢𝐜𝐭𝐢𝐯𝐞 𝐏𝐫𝐨𝐜𝐞𝐬𝐬 𝐌𝐨𝐧𝐢𝐭𝐨𝐫𝐢𝐧𝐠, representing ongoing process executions as natural-language narratives that preserve the relationships among multiple interacting objects.

By combining these representations with a 𝐋𝐚𝐫𝐠𝐞 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥 and a 𝐠𝐥𝐨𝐛𝐚𝐥 𝐥𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐬𝐭𝐫𝐚𝐭𝐞𝐠𝐲, CICERONE can simultaneously generate predictions for all the objects involved in ongoing process executions.

Experiments on benchmark Object-Centric Event Logs demonstrate its effectiveness, while an occlusion-based analysis helps explain the impact of object interactions on the predictions.

🔗 Link to the paper in the first comment below!

25/07/2026

We are pleased to announce that the KDDE research group contributed three papers to the 27th International Symposium on Methodologies for Intelligent Systems — ISMIS 2026.

The papers explore different research challenges in predictive process monitoring, explainable and prescriptive process analytics, online deep learning, time-series forecasting and additive manufacturing.

🔹 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝘃𝗲 𝗣𝗿𝗼𝗰𝗲𝘀𝘀 𝗠𝗼𝗻𝗶𝘁𝗼𝗿𝗶𝗻𝗴 𝗧𝗵𝗿𝗼𝘂𝗴𝗵 𝘁𝗵𝗲 𝗟𝗲𝗻𝘀 𝗼𝗳 𝗗𝗲𝗲𝗽 𝗢𝗻𝗹𝗶𝗻𝗲 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴
Vincenzo Pasquadibisceglie, Simone Capone, Annalisa Appice and Donato Malerba

The paper presents ATLAS, an online learning approach that continuously updates a deep predictive process monitoring model to address changes occurring in evolving business processes.

🔗 https://link.springer.com/chapter/10.1007/978-3-032-32643-0_20

🔹 𝗦𝗶𝗺𝗶𝗹𝗮𝗿𝗶𝘁𝘆-𝗚𝘂𝗶𝗱𝗲𝗱 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝗶𝗻𝗴 𝗼𝗳 𝗠𝘂𝗹𝘁𝗶𝗽𝗹𝗲 𝗦𝗲𝗾𝘂𝗲𝗻𝗰𝗲𝘀 𝗳𝗿𝗼𝗺 𝗠𝗲𝗰𝗵𝗮𝗻𝗶𝗰𝗮𝗹 𝗧𝗲𝘀𝘁𝗶𝗻𝗴 𝗗𝗮𝘁𝗮 𝗼𝗳 𝗔𝗱𝗱𝗶𝘁𝗶𝘃𝗲𝗹𝘆 𝗠𝗮𝗻𝘂𝗳𝗮𝗰𝘁𝘂𝗿𝗲𝗱 𝗖𝗼𝗺𝗽𝗼𝗻𝗲𝗻𝘁𝘀
Anthony Pellicani, Gianvito Pio, Donato Malerba and Michelangelo Ceci

The study introduces a similarity-guided method combining temporal clustering and LSTM-based models to improve multivariate forecasting from heterogeneous mechanical testing data.

🔗 https://link.springer.com/chapter/10.1007/978-3-032-32643-0_18

🏆 🔹 𝗖𝗼𝘂𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝘁𝘂𝗮𝗹𝘀 𝘁𝗼 𝗠𝗮𝗻𝗮𝗴𝗲 𝗢𝗻𝗴𝗼𝗶𝗻𝗴 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗣𝗿𝗼𝗰𝗲𝘀𝘀 𝗗𝗲𝘃𝗶𝗮𝗻𝗰𝗲𝘀
Vincenzo Pasquadibisceglie, Rossella Anna Giansante, Annalisa Appice and Donato Malerba

The paper presents FIREFOX, a methodology that detects potential deviations in ongoing process executions and generates counterfactual recommendations for actions that may prevent undesirable outcomes.

We are especially proud that this contribution received the 𝗜𝗦𝗠𝗜𝗦 𝟮𝟬𝟮𝟲 𝗕𝗲𝘀𝘁 𝗣𝗮𝗽𝗲𝗿 𝗔𝘄𝗮𝗿𝗱. 🏆

🔗 https://link.springer.com/chapter/10.1007/978-3-032-32643-0_12

Congratulations to all the authors on this important collective achievement, which reflects the breadth and quality of the research conducted within the KDDE group!

24/07/2026

🏆 𝗕𝗲𝘀𝘁 𝗣𝗮𝗽𝗲𝗿 𝗔𝘄𝗮𝗿𝗱 𝗮𝘁 𝗜𝗦𝗠𝗜𝗦 𝟮𝟬𝟮𝟲!

We are delighted to announce that the paper

📄 𝗖𝗼𝘂𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝘁𝘂𝗮𝗹𝘀 𝘁𝗼 𝗠𝗮𝗻𝗮𝗴𝗲 𝗢𝗻𝗴𝗼𝗶𝗻𝗴 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗣𝗿𝗼𝗰𝗲𝘀𝘀 𝗗𝗲𝘃𝗶𝗮𝗻𝗰𝗲𝘀

by 𝗩𝗶𝗻𝗰𝗲𝗻𝘇𝗼 𝗣𝗮𝘀𝗾𝘂𝗮𝗱𝗶𝗯𝗶𝘀𝗰𝗲𝗴𝗹𝗶𝗲, 𝗥𝗼𝘀𝘀𝗲𝗹𝗹𝗮 𝗔𝗻𝗻𝗮 𝗚𝗶𝗮𝗻𝘀𝗮𝗻𝘁𝗲, 𝗔𝗻𝗻𝗮𝗹𝗶𝘀𝗮 𝗔𝗽𝗽𝗶𝗰𝗲, and 𝗗𝗼𝗻𝗮𝘁𝗼 𝗠𝗮𝗹𝗲𝗿𝗯𝗮

has received the 𝗕𝗲𝘀𝘁 𝗣𝗮𝗽𝗲𝗿 𝗔𝘄𝗮𝗿𝗱 at the 28th International Symposium on Methodologies for Intelligent Systems — 𝗜𝗦𝗠𝗜𝗦 𝟮𝟬𝟮𝟲, held in Lyon, France.

The paper investigates how counterfactual explanations can support the management of ongoing business process deviations by suggesting actionable changes that may help steer a process towards a desired outcome.

This prestigious recognition rewards the research carried out within the KDDE - Knowledge Discovery and Data Engineering research group at the DIB - Dipartimento di Informatica, Università degli Studi di Bari Aldo Moro.

We are grateful to the ISMIS 2026 organizers and Program Committee for this important recognition and proud to share this achievement with the entire 𝗞𝗗𝗗𝗘 team.

🔗 https://link.springer.com/chapter/10.1007/978-3-032-32643-0_12

25/06/2026

KDDE has organized a seminar for next month.

23/06/2026

🚀 New publication in Data Mining and Knowledge Discovery

We are pleased to announce the publication of the article:

“Dynamic instance weighting for online learning in multi-cryptocurrency price and trend forecasting”

authored by Anthony Pellicani, Gianvito Pio, Sašo Džeroski, and Michelangelo Ceci.

The paper introduces LEMON, a novel online learning approach for real-time forecasting of cryptocurrency price variations and market trends. LEMON leverages temporal correlations among cryptocurrencies, dynamically identifies groups with similar trends, and adopts an adaptive instance-weighting strategy to better model abrupt changes in streaming data.

Experiments on 16 cryptocurrency datasets show that LEMON outperforms state-of-the-art methods in both regression and classification tasks, supporting more accurate real-time predictions in highly volatile markets.

The work is the result of an international collaboration between researchers from the KDDE - Knowledge Discovery and Data Engineering Research Group, Università degli Studi di Bari Aldo Moro , and Sašo Džeroski from the Jožef Stefan Institute.

📖 Read the open access article here:
https://link.springer.com/article/10.1007/s10618-026-01232-9



Springer Nature

The cryptocurrency market represents a significant innovation in the financial ecosystem, built upon cryptographic principles to ensure secure and transparent transactions. Cryptocurrencies experienced a global adoption, driven by their decentralized nature that enables borderless transactions witho...

08/06/2026

We are proud to share the outstanding results achieved by the KDDE - Knowledge Discovery and Data Engineering research group in the latest Italian Research Quality Assessment (VQR).

Although the group was expected to contribute 22.5 research outputs, KDDE submitted 31 outputs, reflecting the strength and productivity of its research activities.

The evaluation results were exceptional: 20 outputs were rated Outstanding and 11 Excellent, meaning that 100% of the submitted research was placed in the two highest quality categories, with almost two-thirds achieving the highest rating.

This achievement is a testament to the scientific excellence, impact, and dedication of all KDDE researchers and collaborators.

Congratulations to everyone who contributed to this remarkable success!

01/05/2026

📢 𝐀 𝐃𝐀𝐓𝐀‑𝐂𝐄𝐍𝐓𝐑𝐈𝐂 𝐕𝐈𝐄𝐖 𝐎𝐍 𝐀𝐑𝐓𝐈𝐅𝐈𝐂𝐈𝐀𝐋 𝐈𝐍𝐓𝐄𝐋𝐋𝐈𝐆𝐄𝐍𝐂𝐄 📢

The KDDE - Knowledge Discovery and Data Engineering is pleased to share a recent research outcome produced within the 𝐅𝐀𝐈𝐑 𝐩𝐫𝐨𝐣𝐞𝐜𝐭, in the framework of 𝐓𝐏𝟕 – 𝐃𝐚𝐭𝐚‑𝐜𝐞𝐧𝐭𝐫𝐢𝐜 𝐀𝐈 𝐚𝐧𝐝 𝐈𝐧𝐟𝐫𝐚𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞𝐬.

An 𝐨𝐩𝐞𝐧‑𝐚𝐜𝐜𝐞𝐬𝐬 𝐬𝐜𝐢𝐞𝐧𝐭𝐢𝐟𝐢𝐜 𝐚𝐫𝐭𝐢𝐜𝐥𝐞 has just been published, evolving a previously released white paper into a full research contribution that serves as a 𝐃𝐚𝐭𝐚‑𝐂𝐞𝐧𝐭𝐫𝐢𝐜 𝐀𝐈 𝐌𝐚𝐧𝐢𝐟𝐞𝐬𝐭𝐨:

👉 https://www.mdpi.com/2079-9292/15/9/1913

The paper advocates a paradigm shift from 𝐦𝐨𝐝𝐞𝐥‑𝐜𝐞𝐧𝐭𝐫𝐢𝐜 𝐀𝐈 to 𝐝𝐚𝐭𝐚‑𝐜𝐞𝐧𝐭𝐫𝐢𝐜 𝐀𝐈. While traditional approaches focus on continuously changing models trained on mostly static datasets, the data‑centric perspective reverses this dynamic:
𝐦𝐨𝐝𝐞𝐥𝐬 𝐛𝐞𝐜𝐨𝐦𝐞 𝐜𝐨𝐦𝐩𝐚𝐫𝐚𝐭𝐢𝐯𝐞𝐥𝐲 𝐬𝐭𝐚𝐛𝐥𝐞, 𝐰𝐡𝐢𝐥𝐞 𝐝𝐚𝐭𝐚 𝐚𝐫𝐞 𝐜𝐨𝐧𝐭𝐢𝐧𝐮𝐨𝐮𝐬𝐥𝐲 𝐜𝐮𝐫𝐚𝐭𝐞𝐝, 𝐞𝐧𝐫𝐢𝐜𝐡𝐞𝐝, 𝐠𝐨𝐯𝐞𝐫𝐧𝐞𝐝, 𝐚𝐧𝐝 𝐢𝐦𝐩𝐫𝐨𝐯𝐞𝐝 throughout the AI lifecycle.

The work provides:
• a 𝐦𝐞𝐭𝐡𝐨𝐝𝐨𝐥𝐨𝐠𝐢𝐜𝐚𝐥 𝐚𝐧𝐝 𝐜𝐨𝐧𝐜𝐞𝐩𝐭𝐮𝐚𝐥 𝐟𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧 for Data‑centric AI;
• a clear 𝐜𝐨𝐧𝐧𝐞𝐜𝐭𝐢𝐨𝐧 𝐰𝐢𝐭𝐡 𝐭𝐨𝐨𝐥𝐬, 𝐢𝐧𝐟𝐫𝐚𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞𝐬, 𝐚𝐧𝐝 𝐅𝐀𝐈𝐑 𝐝𝐚𝐭𝐚 𝐩𝐫𝐢𝐧𝐜𝐢𝐩𝐥𝐞𝐬;
• an up‑to‑date discussion of 𝐃𝐚𝐭𝐚‑𝐜𝐞𝐧𝐭𝐫𝐢𝐜 𝐀𝐈 𝐢𝐧 𝐭𝐡𝐞 𝐞𝐫𝐚 𝐨𝐟 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐯𝐞 𝐀𝐈, with emphasis on robustness, reliability, and responsible deployment.
This contribution highlights how 𝐝𝐚𝐭𝐚 𝐪𝐮𝐚𝐥𝐢𝐭𝐲 𝐚𝐧𝐝 𝐝𝐚𝐭𝐚 𝐩𝐫𝐨𝐜𝐞𝐬𝐬𝐞𝐬 𝐚𝐫𝐞 𝐤𝐞𝐲 𝐝𝐫𝐢𝐯𝐞𝐫𝐬 𝐨𝐟 𝐦𝐨𝐝𝐞𝐫𝐧 𝐀𝐈 𝐬𝐲𝐬𝐭𝐞𝐦𝐬, offering a reference framework for researchers, practitioners, and institutions.

𝐀𝐮𝐭𝐡𝐨𝐫𝐬
Donato Malerba
Antonella Poggi
Mario Alviano
Tommaso Boccali
Maria Teresa Camerlingo
Roberto Maria Delfino
Domenico Diacono
Domenico Elia
Vincenzo Pasquadibisceglie
Mara Sangiovanni
Vincenzo Spinoso
Gioacchino Vino

18/03/2026

🚀 𝗡𝗲𝘄 𝗢𝗽𝗲𝗻 𝗔𝗰𝗰𝗲𝘀𝘀 𝗣𝘂𝗯𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗯𝘆 𝗞𝗗𝗗𝗘 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗚𝗿𝗼𝘂𝗽!

Our new article — authored by 𝗖𝗼𝗿𝗿𝗮𝗱𝗼 𝗟𝗼𝗴𝗹𝗶𝘀𝗰𝗶, 𝗩𝗶𝘁𝗼 𝗡. 𝗟𝗼𝘀𝗮𝘃𝗶𝗼, 𝗦𝗮𝘃𝗲𝗿𝗶𝗼 𝗣𝗮𝘀𝗰𝗮𝘇𝗶𝗼 𝗮𝗻𝗱 𝗗𝗼𝗻𝗮𝘁𝗼 𝗠𝗮𝗹𝗲𝗿𝗯𝗮 — has just been published Open Access in 𝑸𝒖𝒂𝒏𝒕𝒖𝒎 𝑴𝒂𝒄𝒉𝒊𝒏𝒆 𝑰𝒏𝒕𝒆𝒍𝒍𝒊𝒈𝒆𝒏𝒄𝒆 (Springer).

🔗 𝗥𝗲𝗮𝗱 𝘁𝗵𝗲 𝗽𝗮𝗽𝗲𝗿 𝗵𝗲𝗿𝗲:
https://link.springer.com/article/10.1007/s42484-026-00374-9

This work introduces 𝗤𝗨𝗥𝗜𝗢𝗦𝗢, a framework that helps machine learning models become more efficient and reliable when dealing with real‑world, continuously evolving data. Instead of re‑optimizing quantum circuit parameters from scratch each time, QURIOSO predicts good parameters using classical or quantum‑enhanced LSTM models—leading to faster, more stable learning.

Why does this matter?
Because in many real applications, from streaming data to dynamic environments, models need to adapt quickly. Making quantum‑enhanced AI systems more robust and less dependent on costly re‑optimization is a key step toward practical quantum machine learning.

We’re proud to share this contribution with the scientific community.

In Variational Quantum Algorithms (VQAs), circuit parameters are typically re-optimized from scratch for each new dataset, an approach that becomes ineffic

06/03/2026

📢 New press release from the Università degli Studi di Bari Aldo Moro

🌳 An Artificial Intelligence model for satellite image analysis to monitor forest health.

The study proposes an AI-based approach to analyze satellite imagery and support forest monitoring, contributing to ecosystem protection and the sustainable management of natural resources.

🔗 https://www.uniba.it/it/ateneo/rettorato/ufficio-stampa/comunicati-stampa/anno-2026/modello-intelligenza-artificiale-analisi-immagini-satellitari-tato-salute-foreste

The research activity was supported by the SWIFTT Project and the PNRR project FAIR – Future AI Research.

Researchers involved in the study:
Vito Recchia, Giuseppina Andresini, Annalisa Appice, Dino Ienco, Giuseppe Fiameni, Donato Malerba.

5 marzo 2026